No separate experiment setup needed. On the Edit assets page, a new panel called Save and test asset group changes has a toggle for it.
From there you review control and treatment, experiment name, treatment traffic split and duration.
By default, a winning treatment is applied to the original asset group automatically.
Feed-only experiments work differently. You need an existing campaign with at least one paused feed-only asset group.
The experiment type is still in beta.
Spotted by Hana Kobzová, via PPC News Feed.
